←搜索结果

按匹配度 · 共 3407 条结果 清除搜索

星期三

  1. The New Stack

    脚本测试跟不上 AI 写的代码,Datadog 押注智能体测试

    Datadog 今年早些时候收购了 Propolis,将智能体编写、面向目标和输出的测试能力加入其平台,智能体可自行探索应用、跟随用户流程,并根据自然语言提示生成测试。文章称 Anthropic 工程师的代码产出已达此前的 8 倍,部分工程师每月提交数千个 PR,测试成为软件开发生命周期的新瓶颈。Datadog 的 Ruxanda Lueck 与 Propolis 联合创始人 Matthew White 将于 11 月 4 日参与 The New Stack 的线上活动,讨论智能体测试的现状与下一步。

    ain-language prompts. Add Datadog’s observability heft, and you get a system that can test code while providing the logs and traces it needs to vet system performance without the fragility hardcoded testing brings. You have successfully registered for the webinar. The New Stack will sit down with Ruxanda Lueck and Matthew White of Datadog on Wednesday, November 4, at 9 a.m. Pacific / 12 p.m. Eastern. Lueck is a group product manager working on goal-based testing of AI features. White co-founded

    匹配字段正文

  2. The New Stack

    Cloudflare 牵头用 AI 智能体打造的 Rust 版 TypeScript 移植 tsrs

    Cloudflare 接管了由 AI 智能体编写的 Rust 版 TypeScript 移植项目 tsrs,其速度比 TypeScript 7.0 快 8 倍、内存占用减少 70%。该项目由 Owner.com 的 Max Schwenk 用 Claude Opus 5.5 驱动 20 多个子智能体完成,他本人未写一行代码,将 37,000 文件的类型检查从 80 秒降至 CI 中约 7 秒。Cloudflare 强调这不是 fork,而是与 TypeScript 团队共享优化,TypeScript 7.1 已因此提速约 2.5 倍。

    o Rust. Elsewhere, Meta has also used AI to help port the React Compiler to Rust, while in September OpenAI revealed two engineers used Codex and GPT-5.5 to rebuild its Habitat storage service in Rust. But such rapid, AI-assisted rewrites raise questions about how developers verify that the resulting software behaves as intended. For a TypeScript compiler, that means ensuring it catches the same errors as Microsoft’s implementation. Schwenk says the project uses TypeScript’s conformance tests, d

    匹配字段正文

  3. SiliconANGLE

    智能体正在打破 AI 经济性,解法要从数据层开始

    Couchbase 扩展 AI Data Plane,推出原生 Amazon Bedrock 集成与语义缓存、AWS 上全托管的 Agent Memory 和 MCP server,以及分析引擎 Operational Insights 中的向量搜索。Couchbase 称语义缓存可减少 40% 至 60% 的 LLM 调用,Agent Memory 在 LoCoMo 基准上比替代方案少用 33% 至 45% token,向量检索让提示词缩小约 70%。其 AI 高级副总裁 Tim Finley 指出,智能体每任务产生的流量比人类高 450%,客户在规模化后常为账单而降级模型。

    f industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations. LATEST STORIES OpenAI acquires Orchestra Bio to bring AI agents into drug development DeepSeek open-sources DeepInfer, the inference framework behind its production services I

    匹配字段正文

  4. SiliconANGLE

    Zoom 在 Zoomtopia 推出统一上下文层及多项 AI 产品更新

    Zoom 在 Zoomtopia 大会上发布 Unified Context Layer,把会议、通话和客户互动信息与 Salesforce、ServiceNow、Workday 等系统数据整合成实时更新的关系图谱,供 AI 智能体共享记忆和协同工作,该层已可用。Zoom 称内部基准显示这一方式比独立智能体各自解析信息降低成本 45% 至 65%,但未提供独立验证。同期还推出 ZoomMate Agent Teams、扩展到 40 多个第三方应用的智能体搜索、支持多语言广播的翻译与语音克隆,以及通过 MCP 服务器和 Search、Ask API 向外部应用开放上下文,多数功能计划在 2027 年初前陆续发布。

    f industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations. LATEST STORIES OpenAI acquires Orchestra Bio to bring AI agents into drug development DeepSeek open-sources DeepInfer, the inference framework behind its production services I

    匹配字段正文

  5. SiliconANGLE

    Grafana Labs 推出 AI 智能体,实现 AI 智能体可观测性自动化

    Grafana Labs 在 ObservabilityCON 2026 上为 Grafana Cloud 推出 Assistant Watchers 等 AI 智能体,用于监控 AI 工作负载与自主智能体并标记行为异常。同步预览 Grafana Managed Sandboxes,可在受控环境中分析代码并准备供人工审核的修复方案;另有 Pages 提供应用式可视化。Grafana Cloud Logs 借助收购 Logline 所得技术提升查询性能,对低频数据全文搜索不额外收费。

    l cost, Grafana said. Another update is the new Adaptive Logs Archive within Grafana’s Adaptive Telemetry Suite, which allows enterprise users to route logs to their own object storage buckets using open data formats so they can retain observability data at lower cost for future investigations and compliance checks. Finally, Grafana said, it’s trying to address full-stack visibility across both AI and traditional cloud workloads with the introduction of its new Knowledge Graph and Digital Experi

    匹配字段正文

  6. SiliconANGLE

    Dell 首席安全官:AI 采用缓慢是金融转型最大风险

    Dell 总裁兼首席安全官 John Scimone 警告,智能体 AI 大幅降低了发动高级网络攻击的门槛,最不复杂的攻击也比一年前更强。Dell 已参与 Anthropic 的 Project Glasswing,用前沿模型加固自身防御,并主张安全团队必须用 AI 对抗 AI。他认为未来安全团队可能只有一人,却配有 1 万个智能体。

    nsufficient, he explained. “We as cybersecurity professionals are going to have to use AI to defend against AI,” Scimone said. “There’s a huge opportunity really to be able to leverage this capability for good.” That opportunity could matter most to organizations that have never been able to staff a full security team. Agents can deliver scale and expertise that smaller companies could never hire or build on their own, according to Scimone. “Our future security program might be a security progra

    匹配字段正文

  7. SiliconANGLE

    monday.com Elevate 大会前瞻:企业工作管理迈向 AI 智能体执行

    monday.com 将于 10 月 28–29 日在纽约 Javits Center 举办 Elevate 大会,主题是企业工作管理从辅助员工转向由 AI 智能体执行业务流程。其 monday AI 生态包含 Agent Factory、monday sidekick 和 monday vibe,后者支持用自然语言提示词构建自定义业务应用。theCUBE Research 数据显示,92% 的组织已在软件交付生命周期至少一个阶段集成 AI 能力,高于 2024 年初的 71%。

    elopers and technology professionals to examine how AI is changing enterprise operations. SiliconANGLE and theCUBE will provide exclusive interviews and analysis, with special coverage airing Oct. 29. ( Disclosure below.) “Monday.com Elevate signals a broader shift in application development: from building software that helps people manage work to building AI-powered applications and agents that execute work,” said Paul Nashawaty, principal analyst at theCUBE Research. “For AppDev leaders, the p

    匹配字段正文

  8. SiliconANGLE

    Dell CFO 谈金融服务 AI 规模化:先为 token 做预算,再追求速度

    Dell CFO David Kennedy 表示,企业 AI 正从试验走向生产,多数 CFO 已为 token 支出设立专门预算项,但真正的价值驱动是速度与生产力,而非控制成本。Dell 客户中已有 6,500 个 AI factory,5 月还扩展了与 Nvidia 合作的 AI Factory;重复性财务工作流可改用本地开源权重模型,把可变 token 成本转为固定成本。

    ol toward competitive speed. Dell Technologies Inc., fresh off another blowout quarter driven by AI servers, sees demand spreading well beyond GPUs into traditional servers and storage, according to David Kennedy (pictured), chief financial officer of Dell. “The spend piece, honestly, will take care of itself,” Kennedy said. “It needs to be managed. IT and finance can help put guardrails around that. I think it’s important, but the real value driver is the speed and productivity an enterprise ca

    匹配字段正文

  9. SiliconANGLE

    Inception 发布扩散语言模型 Mercury 3,称速度达 GPT-6 Luna 的 8 倍

    Inception AI 发布旗舰扩散语言模型 Mercury 3,在自家 12 项基准中 8 项超过 GPT-6 Luna 和 Claude Haiku 5.5,速度为其 4 到 8 倍。官方称在 NVIDIA GPU 上达到每秒 1068 tokens,Haiku 5.5 约 240、GPT-6 Luna 约 130;上下文窗口提升至 100 万 tokens,参数量未披露,仅称超过 Mercury 1 的 40 倍。定价为每百万输入 tokens 20 美分、输出 70 美分,发布期五折,可通过 Inception 平台及 OpenRouter、Baseten、Vercel 调用。

    f industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations. LATEST STORIES Eka Robotics nets $150M at a $1.45B valuation for robot hands that sense force OpenAI acquires Orchestra Bio to bring AI agents into drug development DeepSeek o

    匹配字段正文

  10. SiliconANGLE

    DeepSeek 开源生产推理框架 DeepInfer

    DeepSeek 开源了支撑其生产服务的推理框架 DeepInfer,采用 Apache 2.0 许可证发布在 GitHub 上。目前该框架只支持 DeepSeek-V4.1-Flash 一个模型,默认加载周末发布的 1017 版本,可运行在 Nvidia SM100 Blackwell GPU 和华为昇腾 950 NPU 上,不支持 Hopper 等较旧的 Nvidia 架构。DeepSeek 称单个集群每天服务 155 万亿 token,并暂未开源与内部基础设施强耦合的 KV cache 管理和负载均衡代码。

    f industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations. LATEST STORIES OpenAI acquires Orchestra Bio to bring AI agents into drug development Inception says Mercury 3 beats GPT-6 Luna and Claude Haiku 5.5 at up to 8x the speed Otte

    匹配字段正文

  11. SiliconANGLE

    OpenAI 收购 Orchestra Bio,将 AI 智能体引入药物研发

    OpenAI 宣布收购药物研发软件初创公司 Orchestra Bio,其团队与技术将并入 OpenAI 的 Rosalind 生命科学部门。Orchestra 的智能体负责监控研究进展、跟踪财务并处理临床试验排期问题,最终决策仍由人做出。交易金额未披露,OpenAI 也未给出这些智能体进入其产品的时间表;Rosalind 业务源于今年 4 月发布的 GPT-Rosalind 生物与药物发现推理模型,诺和诺德与 Moderna 已在合作使用。

    unit. Much of OpenAI’s case for the deal comes down to how long drugs take to reach patients. Getting a new medicine through to U.S. regulatory approval takes 10 to 15 years on average by the company’s count, and programs pass between departments and outside partners whose systems often do not connect. Orchestra’s agents take over the monitoring of studies, and TechCrunch’s write-up of the deal credits them with keeping track of finances and sorting out clinical trial scheduling problems as well

    匹配字段正文

  12. 9to5Mac

    Meta 的 Muse 应用 40 天内在美国下载量突破 1000 万

    Sensor Tower 报告显示,Meta 的 Muse 应用上线 40 天内在美国下载量突破 1000 万,成为自 2015 年 10 月开始追踪以来第 12 个达成这一数字的应用。作为对比,ChatGPT、Grok 和 Claude 在各自上线前 40 天的美国下载量分别为 370 万、50 万和 20 万。Muse 的 7 日留存率为 30%、14 日留存率为 23%,高于 ChatGPT、Claude 和 Grok 平均的 19% 和 16%;本月该应用占 Meta 美国广告支出的 39%,高于 9 月的 23%。报告发布之际,Muse 正因共享对话可能泄露敏感信息而面临隐私争议,Meta 已暂停共享功能并展开调查。

    a better job of retaining users than its AI competitors, which is not surprising, since this is one of Meta’s strongest suits: US Muse users are returning to the app, which is averaging a 30% retention rate after 7 days, and a 23% retention rate after 14 days. Meta is currently outpacing AI competitors in its retention at the 7 and 14 day mark, as ChatGPT, Claude, and Grok averaged a 19% and 16% retention rate, respectively. Finally, Sensor Tower notes that Meta continues spending heavily to pr

    匹配字段正文

  13. Computerworld

    AI 没有制造 IT 数据问题,只是把它暴露了出来

    AI 试点在 IT 运营中表现不佳,根源常是 CMDB 两年未核对、资产数据散落在表格、各团队对同一字段用法不一等数据碎片化问题。Ivanti 2026 年网络安全研究显示,48% 的安全从业者称 IT 团队对安全问题响应不及时,40% 称 IT 不了解组织的风险容忍度。在将 AI 视为业务关键的组织中,54% 的 IT 从业者表示 AI 让工作更快更好,而仍停留在试点阶段的公司这一比例为 24%。

    ase that hasn’t been reconciled in two years, in asset data in spreadsheets, or in a service desk where three teams use the same category field in three different ways. Transform fragmented IT data into the trusted data foundation AI needs → AI doesn’t create these problems. It surfaces them quickly and publicly. The silo problem nobody owns Most IT environments grew one tool at a time. The endpoint team adopted a management platform. Security brought in a scanner. Service management built workf

    匹配字段正文

  14. The Decoder

    OpenAI 称 GPT-6 在 COBOL、Fortran 等遗留编程语言上表现大幅提升

    OpenAI 表示其最新模型针对 COBOL、Fortran、Visual Basic 等低资源语言做了改进,这些语言训练代码远少于 Python,但支撑着全球关键基础设施。OpenAI 构建了两个评测:MultiLangBench 要求模型从零构建兼容 SQLite 的数据库引擎,GPT-6 Astra 在最高推理强度下 20 种语言平均测试通过率 87.4%,GPT-5.6 Terra 约为 38%;GPT-6 Luna 为 40.3%,略高于 GPT-5.6 Terra 的 37.9% 但成本更低。在将 COBOL 转译为 C++、Java、Kotlin、Python 或 Rust 的迁移基准上,GPT-6.1 Sol 低推理强度下准确率 86.6%,GPT-5.6 Sol 为 5.4%,提高推理强度后可达 97.9%。

    aluations to measure progress. MultiLangBench (GitHub) asks models to build a SQLite-compatible database engine from scratch. With maximum reasoning effort, GPT-6 Astra hits an average test pass rate of 87.4 percent across 20 languages, up from about 38 percent for GPT-5.6 Terra. GPT-6 Luna scores 40.3 percent, barely edging out GPT-5.6 Terra (37.9 percent) but at a much lower cost . On the COBOL migration benchmark, which translates COBOL into C++, Java, Kotlin, Python, or Rust, GPT-6.1 Sol rea

    匹配字段正文

  15. Amazon Science

    Amazon 与 AVERI 合作,通过独立第三方评审强化 AI 安全评估

    Amazon 与独立非营利机构 AVERI 合作,对 Amazon 内部安全评估数据集开展第三方评审,检查测试本身的质量而非仅模型输出。AVERI 的独立评审确认了主题覆盖全面、测试方法多样、污染风险低三项优势,同时指出多语言覆盖和多步对抗提示词有待加强。Amazon 据此更新测试集、引入分级难度校准框架、建立季度评审机制,并将多语言提示词扩展到所有风险类别。

    el outputs. AVERI's independent review (conducted on a random sample of prompts, without input from Amazon's evaluation team) validated three strengths — comprehensive topical coverage across documented safety categories, diversity of testing approaches, and low contamination risk — while surfacing improvement areas such as broader multilingual coverage and more multistep adversarial prompts. In response, Amazon made four concrete changes: refreshing the test set with harder adversarial scenario

    匹配字段正文

星期二

  1. SiliconANGLE

    Celigo 推出 MCP 工具管理平台,管控企业 AI 助手与智能体访问

    Celigo 发布 Model Context Protocol Management 平台,让 IT 与安全团队统一管控 AI 助手和智能体在企业系统中的工具调用。平台通过 MCP Gateway 逐个工具审批访问权限,并借助注册表发现所有 MCP 服务器,可连接 1000 多个应用,每次调用都能追溯到真实用户。

    f industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations. LATEST STORIES Eka Robotics nets $150M at a $1.45B valuation for robot hands that sense force OpenAI acquires Orchestra Bio to bring AI agents into drug development DeepSeek o

    匹配字段正文

  2. 9to5Mac

    苹果解释 HomeView 为何直到现在才能发布

    苹果高管向 The Verge 表示,HomeView 之所以长期推迟,是因为其核心卖点依赖 Siri AI 带来的个人上下文与对话能力,而这一能力此前无法交付。HomeView 将于 2026 年 11 月 13 日上市,苹果将其定位为面向家庭共享、可用自然语言交互的设备。由于整体价值围绕新助手展开,HomeView 在欧盟的缺席也因此更易理解。

    t do you think of Apple’s explanation for the HomeView’s launch timing? Let us know in the comments. Best Apple Home accessories eufy 4K camera with HomeKit Secure Video HomeKit Garage Door Opener HomeKit smart plug 4-pack Check out 9to5Mac on YouTube for more Apple news: Ryan got his start in journalism as an Editor at MacStories, where he worked for four years covering Apple news, writing app reviews, and more. For two years he co-hosted the Adapt podcast on Relay FM, which focused entirely on

    匹配字段正文

  3. The Guardian Technology

    科技亿万富翁如何塑造世界:SpaceX 在乌克兰限速与 AI 生成音乐的争议

    SpaceX 的卫星互联网服务上周在乌克兰遭限速,导致飞行速度超过 100mph 的无人机失去连接,多项乌军作战任务失败;路透社称限速请求来自白宫,Musk 否认参与。音乐界方面,A$AP Rocky 与 Grimes 均表示接受 AI 生成音乐但自己不做,而回溯模拟乐器的作品正占据 Billboard 榜单。

    an Salazar, a Chilean activist campaigning to save water in his native country in the face of a wave of AI datacentres, showed how his community had set up a human chatbot to encourage people to prompt less and save water. The idea was simple: a team, ranging from children to town elders, offered human responses to chatbot queries from anywhere in the world for a day. If you want an image of a dog doing yoga, one of them will paint it for you. Want to know what love is? A human will tell you it

    匹配字段正文

  4. Mistral AI精选

    Mistral 发布 Voxtral 3 语音模型家族与语音 Agent SDK

    Mistral 发布 Voxtral 3 语音模型家族,包含支持 50+ 语言的转写模型 Voxtral Small Transcribe 3、支持 35+ 语言和 50+ 方言的合成模型 Voxtral Mini TTS 3,以及基于 LiveKit 的语音 Agent SDK,把听、推理、工具调用和说话串成一条实时流式管线,每轮响应在 1 秒内完成。转写模型支持说话人分离、上下文偏置和轮次结束检测,Mini 版本为 4B 参数,批量转写定价 $0.0035/min;合成模型定价 $13/1M 字符,支持 30 分钟音频的语音克隆。该家族可通过 Mistral 的欧洲托管 API 或自部署使用,语音 Agent API 搭配 Mistral Small 4 推理时定价 $0.04/min。

    back

    ed in production settings with domain-specific vocabulary without the need for fine-tuning. FLEURS 45 Forced vs Inferred Word Error Rate — lower is better Forced Inferred Macro-average WER over the intersection of 45 languages in the FLEURS benchmark supported by all seven models. WER is used for all languages except Japanese, Korean, Thai, and Chinese, which use CER. Lower is better. Forced = language specified; Inferred = language auto-detected. Languages: AR, BN, BG, CA, ZH, CS, DA, NL, EN, E

    匹配字段正文

  5. MarkTechPost

    DeepSeek 开源生产推理框架 DeepInfer,支撑 DeepSeek-V4.1-Flash

    DeepSeek 以 Apache 2.0 协议开源了其生产服务所用的推理框架 DeepInfer,该框架目前只支持 DeepSeek-V4.1-Flash,运行在 NVIDIA SM100 和华为昇腾 950 上。官方称单个集群每天可服务 155 万亿 token,在 131,072 上下文、64 卡解码配置下解码吞吐超过 25,000 tokens/s/GPU。此次发布仅覆盖离线推理,KV cache 系统、P/D 分离和负载均衡未包含在内,DeepSeek 表示这些组件与内部基础设施紧密耦合。

    vLLM, SGLang and TensorRT-LLM? FeatureDeepInfervLLMSGLangTensorRT-LLM MaintainerDeepSeekvLLM projectSGLang projectNVIDIA LicenseApache 2.0Apache 2.0Apache 2.0Apache 2.0 Model coverage1 (DeepSeek-V4.1-Flash)200+ architecturesLLMs, VLMs, diffusionSupport matrix HardwareNVIDIA SM100, Huawei Ascend 950NVIDIA, AMD, Intel, CPUs, plugins incl. Ascend, TPUNVIDIA, AMD, TPU, Intel, Apple, Ascend, Moore ThreadsNVIDIA GPUs Expert parallelismYes, with EPLBYesYesWide-EP Speculative decodingDSpark, confidence

    匹配字段正文